
# imports
import numpy as np
import os
import pandas as pd
%matplotlib inline
import matplotlib.pyplot as plt
import seaborn as sns
from numpy.random import default_rng
from datetime import datetime
from plotly.subplots import make_subplots
import plotly.express as px
import plotly.graph_objects as go
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split, cross_val_score, cross_val_predict, GridSearchCV, RandomizedSearchCV 
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import Perceptron
from sklearn.linear_model import SGDClassifier
from sklearn.metrics import accuracy_score
from sklearn import metrics
from sklearn.metrics import precision_score, recall_score, f1_score, classification_report
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import roc_curve
from numpy import argmax

import warnings
warnings.filterwarnings('ignore')


#Importation des données depuis Github 
general_url = "https://raw.githubusercontent.com/anthonylorendeaux/CESI-IA-CSV/master/general_data.csv"
manager_url = "https://raw.githubusercontent.com/anthonylorendeaux/CESI-IA-CSV/master/manager_survey_data.csv"
employee_url = "https://raw.githubusercontent.com/anthonylorendeaux/CESI-IA-CSV/master/employee_survey_data.csv"
in_time_url = "https://raw.githubusercontent.com/anthonylorendeaux/CESI-IA-CSV/master/in_time.csv"
out_time_url = "https://raw.githubusercontent.com/anthonylorendeaux/CESI-IA-CSV/master/out_time.csv"

#Lecture des csv
general_info_data = pd.read_csv(general_url)
manager_survey_data = pd.read_csv(manager_url)
employee_survey_data = pd.read_csv(employee_url)
in_time_data = pd.read_csv(in_time_url)
out_time_data = pd.read_csv(out_time_url)


general_info_data.head(5)


print('Shape of general_info_data :',general_info_data.shape)
general_info_data.info()

Shape of general_info_data : (4410, 24)
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4410 entries, 0 to 4409
Data columns (total 24 columns):
 #   Column                   Non-Null Count  Dtype  
---  ------                   --------------  -----  
 0   Age                      4410 non-null   int64  
 1   Attrition                4410 non-null   object 
 2   BusinessTravel           4410 non-null   object 
 3   Department               4410 non-null   object 
 4   DistanceFromHome         4410 non-null   int64  
 5   Education                4410 non-null   int64  
 6   EducationField           4410 non-null   object 
 7   EmployeeCount            4410 non-null   int64  
 8   EmployeeID               4410 non-null   int64  
 9   Gender                   4410 non-null   object 
 10  JobLevel                 4410 non-null   int64  
 11  JobRole                  4410 non-null   object 
 12  MaritalStatus            4410 non-null   object 
 13  MonthlyIncome            4410 non-null   int64  
 14  NumCompaniesWorked       4391 non-null   float64
 15  Over18                   4410 non-null   object 
 16  PercentSalaryHike        4410 non-null   int64  
 17  StandardHours            4410 non-null   int64  
 18  StockOptionLevel         4410 non-null   int64  
 19  TotalWorkingYears        4401 non-null   float64
 20  TrainingTimesLastYear    4410 non-null   int64  
 21  YearsAtCompany           4410 non-null   int64  
 22  YearsSinceLastPromotion  4410 non-null   int64  
 23  YearsWithCurrManager     4410 non-null   int64  
dtypes: float64(2), int64(14), object(8)
memory usage: 827.0+ KB


manager_survey_data.head(3)


manager_survey_data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4410 entries, 0 to 4409
Data columns (total 3 columns):
 #   Column             Non-Null Count  Dtype
---  ------             --------------  -----
 0   EmployeeID         4410 non-null   int64
 1   JobInvolvement     4410 non-null   int64
 2   PerformanceRating  4410 non-null   int64
dtypes: int64(3)
memory usage: 103.5 KB


employee_survey_data.head()


employee_survey_data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4410 entries, 0 to 4409
Data columns (total 4 columns):
 #   Column                   Non-Null Count  Dtype  
---  ------                   --------------  -----  
 0   EmployeeID               4410 non-null   int64  
 1   EnvironmentSatisfaction  4385 non-null   float64
 2   JobSatisfaction          4390 non-null   float64
 3   WorkLifeBalance          4372 non-null   float64
dtypes: float64(3), int64(1)
memory usage: 137.9 KB


in_time_data.head(2)


in_time_data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4410 entries, 0 to 4409
Columns: 262 entries, Unnamed: 0 to 2015-12-31
dtypes: float64(12), int64(1), object(249)
memory usage: 8.8+ MB


out_time_data.head(2)


out_time_data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4410 entries, 0 to 4409
Columns: 262 entries, Unnamed: 0 to 2015-12-31
dtypes: float64(12), int64(1), object(249)
memory usage: 8.8+ MB


# Renommage des colonnes sans nom de nos csv in et out
in_time_data.rename(columns={"Unnamed: 0": "EmployeeID"}, inplace=True)
in_time_data.set_index('EmployeeID', inplace=True)
in_time_data
out_time_data.rename(columns={"Unnamed: 0": "EmployeeID"}, inplace=True)
out_time_data.set_index('EmployeeID', inplace=True)

# Suppression des colonnesoù l'employée est absent (valeur NaN)
in_time_data=in_time_data.dropna(axis=1,how='all')
out_time_data=out_time_data.dropna(axis=1,how='all')

# Remplacement des NaN par 0
in_time_data.fillna(0, inplace=True)
out_time_data.fillna(0, inplace=True)

out_time_data.head(3)


for date in in_time_data.columns:
    in_time_data[date]=pd.to_datetime(in_time_data[date])
    out_time_data[date]=pd.to_datetime(out_time_data[date])


time_work_per_day=pd.DataFrame()

cols=in_time_data.columns
for col in cols:
    time_work_per_day[col]=((out_time_data[col] - in_time_data[col]).dt.total_seconds() /3600)

time_work_per_day.head()


time_work_per_day['MeanTimeWorkOverYear2015']=round(time_work_per_day.astype(int).mean(axis=1),2)
time_work_per_day['AbsenceOverYear2015']=(time_work_per_day == 0).astype(int).sum(axis=1)
time_work_per_day = time_work_per_day.drop((time_work_per_day.columns[0:-2]), axis = 1)
time_work_per_day.head()


concat_time_csv = general_info_data.merge(time_work_per_day, on='EmployeeID')
concat_manager_csv = concat_time_csv.merge(manager_survey_data, on='EmployeeID')
temp_concat = concat_manager_csv.merge(employee_survey_data, on='EmployeeID')
temp_concat = temp_concat.set_index('EmployeeID')
temp_concat.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 4410 entries, 1 to 4410
Data columns (total 30 columns):
 #   Column                    Non-Null Count  Dtype  
---  ------                    --------------  -----  
 0   Age                       4410 non-null   int64  
 1   Attrition                 4410 non-null   object 
 2   BusinessTravel            4410 non-null   object 
 3   Department                4410 non-null   object 
 4   DistanceFromHome          4410 non-null   int64  
 5   Education                 4410 non-null   int64  
 6   EducationField            4410 non-null   object 
 7   EmployeeCount             4410 non-null   int64  
 8   Gender                    4410 non-null   object 
 9   JobLevel                  4410 non-null   int64  
 10  JobRole                   4410 non-null   object 
 11  MaritalStatus             4410 non-null   object 
 12  MonthlyIncome             4410 non-null   int64  
 13  NumCompaniesWorked        4391 non-null   float64
 14  Over18                    4410 non-null   object 
 15  PercentSalaryHike         4410 non-null   int64  
 16  StandardHours             4410 non-null   int64  
 17  StockOptionLevel          4410 non-null   int64  
 18  TotalWorkingYears         4401 non-null   float64
 19  TrainingTimesLastYear     4410 non-null   int64  
 20  YearsAtCompany            4410 non-null   int64  
 21  YearsSinceLastPromotion   4410 non-null   int64  
 22  YearsWithCurrManager      4410 non-null   int64  
 23  MeanTimeWorkOverYear2015  4410 non-null   float64
 24  AbsenceOverYear2015       4410 non-null   int64  
 25  JobInvolvement            4410 non-null   int64  
 26  PerformanceRating         4410 non-null   int64  
 27  EnvironmentSatisfaction   4385 non-null   float64
 28  JobSatisfaction           4390 non-null   float64
 29  WorkLifeBalance           4372 non-null   float64
dtypes: float64(6), int64(16), object(8)
memory usage: 1.0+ MB


# Nombre de champs avec des valeurs vides
final_data = temp_concat.copy()
final_data[final_data.columns[final_data.isnull().any()]].isnull().sum()

NumCompaniesWorked         19
TotalWorkingYears           9
EnvironmentSatisfaction    25
JobSatisfaction            20
WorkLifeBalance            38
dtype: int64


# Ajout des valeurs médianes
final_data.fillna(round(final_data.median()),inplace=True)
final_data.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 4410 entries, 1 to 4410
Data columns (total 30 columns):
 #   Column                    Non-Null Count  Dtype  
---  ------                    --------------  -----  
 0   Age                       4410 non-null   int64  
 1   Attrition                 4410 non-null   object 
 2   BusinessTravel            4410 non-null   object 
 3   Department                4410 non-null   object 
 4   DistanceFromHome          4410 non-null   int64  
 5   Education                 4410 non-null   int64  
 6   EducationField            4410 non-null   object 
 7   EmployeeCount             4410 non-null   int64  
 8   Gender                    4410 non-null   object 
 9   JobLevel                  4410 non-null   int64  
 10  JobRole                   4410 non-null   object 
 11  MaritalStatus             4410 non-null   object 
 12  MonthlyIncome             4410 non-null   int64  
 13  NumCompaniesWorked        4410 non-null   float64
 14  Over18                    4410 non-null   object 
 15  PercentSalaryHike         4410 non-null   int64  
 16  StandardHours             4410 non-null   int64  
 17  StockOptionLevel          4410 non-null   int64  
 18  TotalWorkingYears         4410 non-null   float64
 19  TrainingTimesLastYear     4410 non-null   int64  
 20  YearsAtCompany            4410 non-null   int64  
 21  YearsSinceLastPromotion   4410 non-null   int64  
 22  YearsWithCurrManager      4410 non-null   int64  
 23  MeanTimeWorkOverYear2015  4410 non-null   float64
 24  AbsenceOverYear2015       4410 non-null   int64  
 25  JobInvolvement            4410 non-null   int64  
 26  PerformanceRating         4410 non-null   int64  
 27  EnvironmentSatisfaction   4410 non-null   float64
 28  JobSatisfaction           4410 non-null   float64
 29  WorkLifeBalance           4410 non-null   float64
dtypes: float64(6), int64(16), object(8)
memory usage: 1.0+ MB


final_data.describe(include='all').T


final_data.drop(['EmployeeCount', 'Over18', 'StandardHours'], axis=1, inplace = True)
final_data.head(5)


# One-hot Encoding
final_data2 = pd.get_dummies(final_data, 
prefix=["BusinessTravel","Department","EducationField", "JobRole","MaritalStatus", "Gender"],
columns=["BusinessTravel","Department","EducationField", "JobRole","MaritalStatus", "Gender"])
final_data2.head(5)


# Suppression de la colonne Attrition
final_data3 = final_data2.drop("Attrition",axis=1)

# Normalisation des données en utilisant le z-score 
norm = StandardScaler().fit_transform(final_data3)
final_data3 = pd.DataFrame(norm, columns=final_data3.columns)
final_data3


# Suppression cols
cols_pas_ethic = ["Gender_Male", "Gender_Female", "YearsWithCurrManager", "PerformanceRating", "Age", "MaritalStatus_Single", "MaritalStatus_Married", "MaritalStatus_Divorced"]
cols_big_corr = ["Department_Sales", "BusinessTravel_Travel_Rarely"]

final_data4 = final_data3.drop(cols_pas_ethic,axis=1)
final_data4 = final_data4.drop(cols_big_corr,axis=1)
final_data4.head()


# Définitions de nos variables X et y
X = final_data4
y = final_data2["Attrition"]


seed =42
test_size = 0.2
# Fixer un seed pour avoir les mêmes résultats à chaque essai
np.random.seed(seed)

# Séparer les données d'entrainement et de test
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=test_size)


def area_chart(df,feature):
    ax=sns.kdeplot(df[df['Attrition']=='Yes'][feature],
             shade=True,label='Attrition = Yes')
    ax=sns.kdeplot(df[df['Attrition']=='No'][feature],
                 shade=True,label='Attrition = No')
    ax.legend()

cols = ['DistanceFromHome','MonthlyIncome','TotalWorkingYears','TrainingTimesLastYear', 'YearsAtCompany', 'YearsSinceLastPromotion','PercentSalaryHike','YearsWithCurrManager','MeanTimeWorkOverYear2015','AbsenceOverYear2015', "Age"]

plt.figure(figsize=(40,30))
for i, col in enumerate(cols):
    ax = plt.subplot(4, 3, i+1)
    area_chart(final_data, col)
plt.tight_layout(pad=3.0)
plt.show()


cols = ['DistanceFromHome','MonthlyIncome','TotalWorkingYears','TrainingTimesLastYear', 'YearsAtCompany', 'YearsSinceLastPromotion','PercentSalaryHike','YearsWithCurrManager','MeanTimeWorkOverYear2015','AbsenceOverYear2015']

plot_rows=2
plot_cols=5
fig = make_subplots(rows=plot_rows, cols=plot_cols,subplot_titles=(cols))

col_i = 0
for i in range(1, plot_rows + 1):
    for j in range(1, plot_cols + 1):
        for t in px.box(final_data, x="Attrition", y=cols[col_i]).data:
            fig.add_trace(t,row=i, col=j)

        col_i=col_i+1
fig.update_layout(height=600, width=1200)
fig.show()


cols = ['BusinessTravel', 'Department', 'Education', 'EducationField', 'Gender', 'JobLevel', 'JobRole', 'MaritalStatus', 'StockOptionLevel', 'JobInvolvement', 'PerformanceRating', 'EnvironmentSatisfaction', 'JobSatisfaction', 'WorkLifeBalance']

attrition = final_data['Attrition'].tolist()
result = {}

for col in cols:
    current_col = final_data[col].tolist()

    #values = {name: "", [0,1]}
    values = {}
    label = []
    list_yes = []
    list_no = []

    for index in range(len(current_col)):

        try:
            i = label.index(current_col[index])
        except:
            i = -1
        
        # On ajoute
        if(i != -1):
            if(attrition[index] == 'Yes'):
                list_yes[i] += 1
            else:
                list_no[i] += 1
        # On créée
        else:
            label.append(current_col[index])
            if(attrition[index] == 'Yes'):
                list_yes.append(1)
                list_no.append(0)
            else:
                list_yes.append(0)
                list_no.append(1)
    
    pourcent_yes = []
    pourcent_no = []

    for j in range(len(list_yes)):
        no = list_no[j]
        yes = list_yes[j]
        total = no + yes
        pourcent_yes.append("{}%".format(round(100 * yes / total)))
        pourcent_no.append("{}%".format(round(100 * no / total)))
    
    plot_rows=7
    plot_cols=2
    fig = go.Figure()

    fig.add_trace(go.Bar(
                        name='Yes',
                        x=label,
                        y=list_yes,
                        offsetgroup=0,
                        text = pourcent_yes,
                        marker_color="steelblue",
                    ))
    fig.add_trace(go.Bar(
                        name='No',
                        x=label,
                        y=list_no,
                        offsetgroup=1,
                        text = pourcent_no,
                        marker_color="coral",
                    ))
    fig.update_layout(height=500, width=800, title="Attrition for {}".format(col))
    fig.show()


final_data.hist(bins=50, figsize=(20,15))
plt.show()


# Matrice de corrélation
corr_matrix = final_data2.corr(method="pearson")

plt.figure(figsize=(40,30))
sns.heatmap(corr_matrix, annot= True, cbar=True, cmap="Blues_r")
plt.show()

corr_matrix = corr_matrix.where(np.triu(np.ones(corr_matrix.shape), k=1).astype(bool))
corr_matrix = corr_matrix.unstack().reset_index()
corr_matrix.columns = ["feature1", "feature2", "Correlation"]
corr_matrix.dropna(subset = ['Correlation'], inplace = True)
corr_matrix['Correlation'] = round(corr_matrix['Correlation'], 2)
corr_matrix['Correlation'] = abs(corr_matrix['Correlation'])
corr_matrix = corr_matrix.sort_values(by = 'Correlation', ascending = False)
value_high_corr = corr_matrix[corr_matrix['Correlation']>0.7]

value_high_corr


# Creation d'un jeu d'entrainement labelisé
y_train_label = y_train.copy()
y_train_label.where(y_train_label == "Yes", 0, inplace=True)
y_train_label.where(y_train_label == 0, 1, inplace=True)
y_train_label = y_train_label.astype(np.int64)

# Creation d'un jeu de test labelisé
y_test_label = y_test.copy()
y_test_label.where(y_test_label == "Yes", 0, inplace=True)
y_test_label.where(y_test_label == 0, 1, inplace=True)
y_test_label = y_test_label.astype(np.int64)


# Listes des algorithmes  
models = {"Logistic Regression": LogisticRegression(),
          "KNN": KNeighborsClassifier(),
          "Random Forest": RandomForestClassifier(),
          "Perceptron": Perceptron(),
          "Descente de gradient": SGDClassifier()}

# Fonction appliquant un fit et score sur chacun des modeles
def fit_and_score(models, X_train, y_train_label):
    
    # Fixer un seed pour avoir les mêmes résultats à chaque essai 
    np.random.seed(42)
    
    # Dictionnaire pour sauvegarder les scores
    fitted_model = {}
    for name, model in models.items():
        # Fit le modèle
        model.fit(X_train, y_train_label)
        fitted_model[name] = model
        
    return fitted_model


# Entrainement de nos algorithmes
fitted_model = fit_and_score(models=models, X_train=X_train, y_train_label=y_train_label)
fitted_model

{'Logistic Regression': LogisticRegression(),
 'KNN': KNeighborsClassifier(),
 'Random Forest': RandomForestClassifier(),
 'Perceptron': Perceptron(),
 'Descente de gradient': SGDClassifier()}


 def recall_precision(estimator_train, X_train, y_true_train):
    # Print les infos de base (précision, recall, f1...)
    y_pred_train = cross_val_predict(estimator_train, X_train, y_true_train, cv=7)
    precision_train = precision_score(y_true_train, y_pred_train)
    recall_train = recall_score(y_true_train, y_pred_train)
    f1_train = f1_score(y_true_train, y_pred_train)
        
    return precision_train, recall_train, f1_train


# Jeu d'entrainement 
precision_reglog, recall_reglog, f1_reglog = recall_precision(fitted_model['Logistic Regression'], X_train, y_train_label)
precision_knn, recall_knn, f1_knn = recall_precision(fitted_model['KNN'], X_train, y_train_label)
precision_rf, recall_rf, f1_rf = recall_precision(fitted_model['Random Forest'], X_train, y_train_label)
precision_perceptron, recall_perceptron, f1_perceptron = recall_precision(fitted_model['Perceptron'], X_train, y_train_label)
precision_dgs, recall_dgs, f1_dgs = recall_precision(fitted_model['Descente de gradient'], X_train, y_train_label)

data_param = {
    'Algorithmes': ["Logisitic Regression", "KNN", "RandomForest", "Perceptron", "Descente de gradient stochastique"],
    'precision_train': [precision_reglog, precision_knn, precision_rf, precision_perceptron, precision_dgs],
    'recall_train': [recall_reglog, recall_knn, recall_rf, recall_perceptron, recall_dgs],
    'f1_train': [f1_reglog, f1_knn, f1_rf, f1_perceptron, f1_dgs]
}
hyperparameter = pd.DataFrame(data=data_param)
hyperparameter


#Données de paramètres

param_reglog = [
    {'solver': ['newton-cg', 'liblinear'],
    'penalty': ['none', 'l1', 'l2']}
]

param_knn = [
    {'n_neighbors' : range(1,10),
    'weights' : ['uniform', 'distance'],
    'metric' : ['euclidean', 'manhattan']}
]

param_random_forest = [
    {'n_estimators' : [10, 50, 100, 200],
    'max_depth' : [10, 20, 40, 50],
    'max_features': [1, 'auto', 'log2'],
    'bootstrap': [True, False]}
]

param_perceptron = [
    {'penalty' : ['l1', 'l2'],
    'fit_intercept': [True, False],
    'max_iter': [20, 50, 70, 100]}
]

param_descente = [
    {'learning_rate': ['optimal'],
    'early_stopping': [True],
    'validation_fraction': [0.1, 0.3, 0.5],
    'loss': ['log', 'perceptron'],
    'penalty': ['l1', 'l2']}
]


# Fonction calculant la meilleure prediction avec les meilleurs paramètres
def fit_method(gs_cv, X, y):
    gs_cv.fit(X, y)
    gs_cv_best_score = gs_cv.best_score_
    gs_cv_best_param = gs_cv.best_params_
    gs_cv_best_estimator = gs_cv.best_estimator_
    gs_cv_score = cross_val_score(gs_cv_best_estimator, X, y, cv=10).mean()
    
    return gs_cv_best_score, gs_cv_best_param, gs_cv_best_estimator, gs_cv_score


# Grid Search
gs_cv_reglog = GridSearchCV(LogisticRegression(), param_reglog, cv=10,return_train_score=False, verbose=0)

# Jeu de test
reglog_best_score_test, reglog_best_param_test, reglog_best_estimator_test, cv_score_reglog_test =  fit_method(gs_cv_reglog, X_test, y_test_label)
print('reglog_best_score_test:',reglog_best_score_test)

# Jeu d'entrainement
reglog_best_score_train, reglog_best_param_train, reglog_best_estimator_train, cv_score_reglog_train =  fit_method(gs_cv_reglog, X_train, y_train_label)
print('reglog_best_score_train:',reglog_best_score_train)

reglog_best_score_test: 0.8492339121552606
reglog_best_score_train: 0.8528900012876642


# Grid Search
gs_cv_knn = GridSearchCV(KNeighborsClassifier(), param_knn, cv=10,return_train_score=False, verbose=0)

# Jeu de test
knn_best_score_test, knn_best_param_test, knn_best_estimator_test, cv_score_knn_test =  fit_method(gs_cv_knn, X_test, y_test_label)
print('knn_best_score_test:', knn_best_score_test)

# Jeu d'entrainement
knn_best_score_train, knn_best_param_train, knn_best_estimator_train, cv_score_knn_train =  fit_method(gs_cv_knn, X_train, y_train_label)
print('knn_best_score_train:', knn_best_score_train)

knn_best_score_test: 0.8855081716036771
knn_best_score_train: 0.9846912825135204


# Grid Search
gs_cv_rf = GridSearchCV(RandomForestClassifier(), param_random_forest, cv=10,return_train_score=False, verbose=0, n_jobs=-1)

# Jeu de test
rf_best_score_test, rf_best_param_test, rf_best_estimator_test, cv_score_rf_test =  fit_method(gs_cv_rf, X_test, y_test_label)
print('rf_best_score_test:', rf_best_score_test)

# Jeu d'entrainement
rf_best_score_train, rf_best_param_train, rf_best_estimator_train, cv_score_rf_train =  fit_method(gs_cv_rf, X_train, y_train_label)
print('rf_best_score_train:', rf_best_score_train)

rf_best_score_test: 0.8956843718079673
rf_best_score_train: 0.9889445982487766


# Grid Search
gs_cv_perceptron = GridSearchCV(Perceptron(), param_perceptron, cv=10,return_train_score=False, verbose=0)

# Jeu de test
perceptron_best_score_test, perceptron_best_param_test, perceptron_best_estimator_test, cv_score_perceptron_test =  fit_method(gs_cv_perceptron, X_test, y_test_label)
print('perceptron_best_score_test:', perceptron_best_score_test)

# Jeu d'entrainement
perceptron_best_score_train, perceptron_best_param_train, perceptron_best_estimator_train, cv_score_perceptron_train =  fit_method(gs_cv_perceptron, X_train, y_train_label)
print('perceptron_best_score_train:', perceptron_best_score_train)

perceptron_best_score_test: 0.7969994892747702
perceptron_best_score_train: 0.7964597283028587


# Grid Search
gs_cv_dgs = GridSearchCV(SGDClassifier(), param_descente, cv=10,return_train_score=False, verbose=0)

# Jeu de test
dgs_best_score_test, dgs_best_param_test, dgs_best_estimator_test, cv_score_dgs_test =  fit_method(gs_cv_dgs, X_test, y_test_label)
print('dgs_best_score_test:', dgs_best_score_test)

# Jeu d'entrainement
dgs_best_score_train, dgs_best_param_train, dgs_best_estimator_train, cv_score_dgs_train =  fit_method(gs_cv_dgs, X_train, y_train_label)
print('dgs_best_score_train:', dgs_best_score_train)

dgs_best_score_test: 0.8106358529111338
dgs_best_score_train: 0.8072471349472057


data_param = {
    'Algorithmes': ["Logisitic Regression", "KNN", "RandomForest", "Perceptron", "Descente de gradient stochastique"],
    'Best_param': [reglog_best_param_train, knn_best_param_train, rf_best_param_train, perceptron_best_param_train, dgs_best_param_train],
    'Best_estimator': [reglog_best_estimator_train, knn_best_estimator_train, rf_best_estimator_train, perceptron_best_estimator_train, dgs_best_estimator_train],
    'Cross-Validation score Train': [cv_score_reglog_train, cv_score_knn_train, cv_score_rf_train, cv_score_perceptron_train, cv_score_dgs_train]
}
hyperparameter = pd.DataFrame(data=data_param)
hyperparameter


# On définie ici nos fonctions pour plot et de transfo
# Convertir une liste avec des 'No' & 'Yes' en 0 & 1 liste
def transform_binary(table):
    y_pred_binary = []
    
    for item in table:
        if item == 'No':
            y_pred_binary.append(0)
        else:
            y_pred_binary.append(1)
    return y_pred_binary

# Plot courbes de précision et recall vs Threshold
def plot_precision_recall_vs_threshold(precisions, recalls, thresholds):
    plt.plot(thresholds, precisions[:-1], "b-", label="Precision", linewidth=2)
    plt.plot(thresholds, recalls[:-1], "g-", label="Recall", linewidth=2)
    plt.xlabel("Threshold", fontsize=16)
    plt.legend(loc="upper left", fontsize=16)
    plt.ylim([0, 1])

# Plot courbes de précision et recall
def plot_precision_vs_recall(precisions, recalls):
    plt.plot(recalls, precisions, "k-", linewidth=2)
    plt.xlabel("Recall", fontsize=16)
    plt.ylabel("Precision", fontsize=16)
    plt.axis([0, 1, 0, 1])

# Plot de la courbe de Roc
def plot_roc_curve(fpr, tpr, label=None):
    roc_auc = metrics.auc(fpr, tpr)
    plt.plot(fpr, tpr, linewidth=2, label=label)
    plt.plot([0, 1], [0, 1], 'k--')
    plt.axis([0, 1, 0, 1])
    plt.xlabel('False Positive Rate', fontsize=16)
    plt.ylabel('True Positive Rate', fontsize=16)


# Affiche les informations pour un modèle donnée
def displayInfo_by_model(estimator_train, estimator_test, X_train, y_true_train, X_test, y_true_test, method, xlim_x, xlim_y, ylim_x, ylim_y ):
    # Fix random
    np.random.seed(42)

    # Print les infos de base (précision, recall, f1...)
    y_pred_train = cross_val_predict(estimator_train, X_train, y_true_train, cv=7)
    precision_train = precision_score(y_true_train, y_pred_train)
    recall_train = recall_score(y_true_train, y_pred_train)
    f1_train = f1_score(y_true_train, y_pred_train)
    
    y_pred_test = cross_val_predict(estimator_test, X_test, y_true_test, cv=7)
    precision_test = precision_score(y_true_test, y_pred_test)
    recall_test = recall_score(y_true_test, y_pred_test)
    f1_test =  f1_score(y_true_test, y_pred_test)
    
    data_train_test = {
    'Analyse': ["precision_score", "recall_score", "f1_score"],
    "Jeu d'entrainement": [precision_train, recall_train, f1_train],
    "Jeu de test": [precision_test, recall_test, f1_test],
    'Différence': [precision_train - precision_test, recall_train - recall_test, f1_train - f1_test]
    }
        
    result_train_test = pd.DataFrame(data=data_train_test)
    display(result_train_test)
    
    # matrice de confusion
    print("Matrice de confusion jeu d'entrainement:\n", metrics.confusion_matrix(y_true_train, y_pred_train))
    print("Matrice de confusion jeu de test:\n", metrics.confusion_matrix(y_true_test, y_pred_test))
    
    # Afficher courbes
    y_scores = cross_val_predict(estimator_train, X_train, y_true_train, cv=7,method=method)
    if(method == 'predict' or method == 'predict_proba'):
        y_scores = y_scores[:,1]
    precisions, recalls, thresholds = precision_recall_curve(y_true_train, y_scores)

    plt.figure(figsize=(8, 4))
    plot_precision_recall_vs_threshold(precisions, recalls, thresholds)
    plt.xlim([xlim_x,xlim_y])
    plt.ylim([ylim_x,ylim_y])
    plt.show()

    plt.figure(figsize=(8, 6))
    plot_precision_vs_recall(precisions, recalls)
    plt.show()   

    fscore = (2 * precisions * recalls) / (precisions + recalls)
    fpr, tpr, thresholds = roc_curve(y_true_train, y_scores)
    plt.figure(figsize=(8, 6))
    plot_roc_curve(fpr, tpr)
    plt.show()
    print('AUC = ', metrics.auc(fpr, tpr))
    
    return fpr, tpr, fscore, thresholds


# creation d'un jeu d'entrainement labélisé
y_test_label = y_test.copy()
y_test_label.where(y_test_label == "Yes", 0, inplace=True)
y_test_label.where(y_test_label == 0, 1, inplace=True)
y_test_label = y_test_label.astype(np.int64)
y_test_label_true = (y_test_label == 1)

# creation d'un jeu de test labélisé
y_train_label = y_train.copy()
y_train_label.where(y_train_label == "Yes", 0, inplace=True)
y_train_label.where(y_train_label == 0, 1, inplace=True)
y_train_label = y_train_label.astype(np.int64)
y_train_label = (y_train_label == 1)


fpr_reglog, tpr_reglog, fscore_reglog, thresholds_reglog = displayInfo_by_model(reglog_best_estimator_train, reglog_best_estimator_test, X_train, y_train_label, X_test, y_test_label_true,"decision_function",-2,0,0,1)

Matrice de confusion jeu d'entrainement:
 [[2881   77]
 [ 448  122]]
Matrice de confusion jeu de test:
 [[720  21]
 [113  28]]

AUC =  0.793238674780257


fpr_knn, tpr_knn, fscore_knn, thresholds_knn = displayInfo_by_model(knn_best_estimator_train, knn_best_estimator_test, X_train, y_train_label, X_test, y_test_label_true,"predict_proba", 0.25,0.45,0.8,1)

Matrice de confusion jeu d'entrainement:
 [[2937   21]
 [  54  516]]
Matrice de confusion jeu de test:
 [[738   3]
 [ 98  43]]

AUC =  0.9608946300843386


fpr_rf, tpr_rf, fscore_rf, thresholds_rf = displayInfo_by_model(rf_best_estimator_train, rf_best_estimator_test, X_train, y_train_label, X_test, y_test_label_true,"predict_proba", 0.1,0.6,0.8,1)

Matrice de confusion jeu d'entrainement:
 [[2956    2]
 [  64  506]]
Matrice de confusion jeu de test:
 [[726  15]
 [ 90  51]]

AUC =  0.9976400602588282


fpr_perceptron, tpr_perceptron, fscore_perceptron, thresholds_perceptron = displayInfo_by_model(perceptron_best_estimator_train, perceptron_best_estimator_test, X_train, y_train_label, X_test, y_test_label_true,"decision_function", -10,10, 0, 1)

Matrice de confusion jeu d'entrainement:
 [[2595  363]
 [ 384  186]]
Matrice de confusion jeu de test:
 [[667  74]
 [ 98  43]]

AUC =  0.6914783578283097


fpr_dgs, tpr_dgs, fscore_dgs, thresholds_dgs = displayInfo_by_model(dgs_best_estimator_train, dgs_best_estimator_test, X_train, y_train_label, X_test, y_test_label_true,"decision_function", -2, 0, 0,1)

Matrice de confusion jeu d'entrainement:
 [[2596  362]
 [ 403  167]]
Matrice de confusion jeu de test:
 [[659  82]
 [101  40]]

AUC =  0.7039280927131893


plt.figure(figsize=(8, 6))
plot_roc_curve(fpr_rf, tpr_rf, "Random Forest")
plot_roc_curve(fpr_knn, tpr_knn, "KNN")
plot_roc_curve(fpr_reglog, tpr_reglog, "Logistic Regression")
plot_roc_curve(fpr_perceptron, tpr_perceptron, "Perceptron")
plot_roc_curve(fpr_dgs, tpr_dgs, "Descente de Gradient Stochastique")
plt.legend(loc="lower right", fontsize=16)
plt.show()


def percentage(unique, counts):
    return (counts[1]/(counts[0]+counts[1]))*100

ix = argmax(fscore_rf)

final_model = rf_best_estimator_train
final_threshold = thresholds_rf[ix]

final_prediction = final_model.predict_proba(X_train)[:, 1]
final_prediction_set = np.where(final_prediction < final_threshold, 0, 1)

unique, counts = np.unique(final_prediction_set, return_counts=True)
dict(zip(unique, counts))

{0: 2958, 1: 570}


percentage(unique, counts)

16.156462585034014


rf_best_estimator_train.feature_importances_

array([0.06054397, 0.03631924, 0.03679454, 0.06849075, 0.04939129,
       0.05667502, 0.03054851, 0.08114009, 0.04065411, 0.07079823,
       0.04208784, 0.08023756, 0.03170984, 0.03171979, 0.04242946,
       0.04284586, 0.03539624, 0.00845799, 0.0184011 , 0.00799763,
       0.01377548, 0.00547447, 0.01234225, 0.00598787, 0.01098843,
       0.00469038, 0.0064952 , 0.00707586, 0.00347438, 0.00965039,
       0.0050706 , 0.00664056, 0.00821099, 0.01067405, 0.01107376,
       0.0057363 ])


feature_dict = dict(sorted(zip(X.columns, list(rf_best_estimator_train.feature_importances_)), reverse=True))
feature_dict_sorted_A = sorted(feature_dict.items(), key=lambda x:x[1], reverse=True)
feature_dict_sorted_A

[('TotalWorkingYears', 0.08114009468198877),
 ('MeanTimeWorkOverYear2015', 0.08023756249134414),
 ('YearsAtCompany', 0.07079822541857282),
 ('MonthlyIncome', 0.06849074838433365),
 ('DistanceFromHome', 0.060543968702801924),
 ('PercentSalaryHike', 0.056675022935297796),
 ('NumCompaniesWorked', 0.04939129332057729),
 ('JobSatisfaction', 0.0428458565509191),
 ('EnvironmentSatisfaction', 0.042429461818388345),
 ('YearsSinceLastPromotion', 0.042087836910063306),
 ('TrainingTimesLastYear', 0.04065410856521336),
 ('JobLevel', 0.03679453735646233),
 ('Education', 0.036319239744907875),
 ('WorkLifeBalance', 0.03539624352822322),
 ('JobInvolvement', 0.03171978549858108),
 ('AbsenceOverYear2015', 0.03170983755884846),
 ('StockOptionLevel', 0.030548511701213324),
 ('BusinessTravel_Travel_Frequently', 0.0184010972339067),
 ('Department_Research & Development', 0.01377547572918943),
 ('EducationField_Life Sciences', 0.012342247951282501),
 ('JobRole_Sales Executive', 0.011073759756276495),
 ('EducationField_Medical', 0.010988433520573007),
 ('JobRole_Research Scientist', 0.010674045255340294),
 ('JobRole_Laboratory Technician', 0.009650393927462059),
 ('BusinessTravel_Non-Travel', 0.00845798604309263),
 ('JobRole_Research Director', 0.008210986124722581),
 ('Department_Human Resources', 0.007997628175530857),
 ('JobRole_Healthcare Representative', 0.007075856579685702),
 ('JobRole_Manufacturing Director', 0.006640557544248407),
 ('EducationField_Technical Degree', 0.0064951975471258016),
 ('EducationField_Marketing', 0.005987866769523813),
 ('JobRole_Sales Representative', 0.005736304367097636),
 ('EducationField_Human Resources', 0.005474466264500532),
 ('JobRole_Manager', 0.005070604546244111),
 ('EducationField_Other', 0.004690376327115722),
 ('JobRole_Human Resources', 0.0034743811693449004)]


feature_df = pd.DataFrame(feature_dict, index=[0])
feature_df.T.plot.barh(figsize=(20,12),title="Feature Importance", legend=False);


area_chart(final_data, 'MeanTimeWorkOverYear2015')


area_chart(final_data, 'MonthlyIncome')


pd.crosstab(final_data["DistanceFromHome"], final_data.Attrition).plot(kind="bar",figsize=(10, 6))
plt.title("Attrition en fonction de DistanceFromHome")
plt.xlabel("DistanceFromHome")
plt.ylabel("F")
plt.legend(["No", "Yes"])
plt.xticks(rotation=0)

(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
        17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28]),
 [Text(0, 0, '1'),
  Text(1, 0, '2'),
  Text(2, 0, '3'),
  Text(3, 0, '4'),
  Text(4, 0, '5'),
  Text(5, 0, '6'),
  Text(6, 0, '7'),
  Text(7, 0, '8'),
  Text(8, 0, '9'),
  Text(9, 0, '10'),
  Text(10, 0, '11'),
  Text(11, 0, '12'),
  Text(12, 0, '13'),
  Text(13, 0, '14'),
  Text(14, 0, '15'),
  Text(15, 0, '16'),
  Text(16, 0, '17'),
  Text(17, 0, '18'),
  Text(18, 0, '19'),
  Text(19, 0, '20'),
  Text(20, 0, '21'),
  Text(21, 0, '22'),
  Text(22, 0, '23'),
  Text(23, 0, '24'),
  Text(24, 0, '25'),
  Text(25, 0, '26'),
  Text(26, 0, '27'),
  Text(27, 0, '28'),
  Text(28, 0, '29')])

	Unnamed: 0	2015-01-01	2015-01-02	2015-01-05	2015-01-06	2015-01-07	2015-01-08	2015-01-09	2015-01-12	2015-01-13	...	2015-12-18	2015-12-21	2015-12-22	2015-12-23	2015-12-24	2015-12-25	2015-12-28	2015-12-29	2015-12-30	2015-12-31
0	1	NaN	2015-01-02 09:43:45	2015-01-05 10:08:48	2015-01-06 09:54:26	2015-01-07 09:34:31	2015-01-08 09:51:09	2015-01-09 10:09:25	2015-01-12 09:42:53	2015-01-13 10:13:06	...	NaN	2015-12-21 09:55:29	2015-12-22 10:04:06	2015-12-23 10:14:27	2015-12-24 10:11:35	NaN	2015-12-28 10:13:41	2015-12-29 10:03:36	2015-12-30 09:54:12	2015-12-31 10:12:44
1	2	NaN	2015-01-02 10:15:44	2015-01-05 10:21:05	NaN	2015-01-07 09:45:17	2015-01-08 10:09:04	2015-01-09 09:43:26	2015-01-12 10:00:07	2015-01-13 10:43:29	...	2015-12-18 10:37:17	2015-12-21 09:49:02	2015-12-22 10:33:51	2015-12-23 10:12:10	NaN	NaN	2015-12-28 09:31:45	2015-12-29 09:55:49	2015-12-30 10:32:25	2015-12-31 09:27:20

	Unnamed: 0	2015-01-01	2015-01-02	2015-01-05	2015-01-06	2015-01-07	2015-01-08	2015-01-09	2015-01-12	2015-01-13	...	2015-12-18	2015-12-21	2015-12-22	2015-12-23	2015-12-24	2015-12-25	2015-12-28	2015-12-29	2015-12-30	2015-12-31
0	1	NaN	2015-01-02 16:56:15	2015-01-05 17:20:11	2015-01-06 17:19:05	2015-01-07 16:34:55	2015-01-08 17:08:32	2015-01-09 17:38:29	2015-01-12 16:58:39	2015-01-13 18:02:58	...	NaN	2015-12-21 17:15:50	2015-12-22 17:27:51	2015-12-23 16:44:44	2015-12-24 17:47:22	NaN	2015-12-28 18:00:07	2015-12-29 17:22:30	2015-12-30 17:40:56	2015-12-31 17:17:33
1	2	NaN	2015-01-02 18:22:17	2015-01-05 17:48:22	NaN	2015-01-07 17:09:06	2015-01-08 17:34:04	2015-01-09 16:52:29	2015-01-12 17:36:48	2015-01-13 18:00:13	...	2015-12-18 18:31:28	2015-12-21 17:34:16	2015-12-22 18:16:35	2015-12-23 17:38:18	NaN	NaN	2015-12-28 17:08:38	2015-12-29 17:54:46	2015-12-30 18:31:35	2015-12-31 17:40:58

	2015-01-02	2015-01-05	2015-01-06	2015-01-07	2015-01-08	2015-01-09	2015-01-12	2015-01-13	2015-01-15	2015-01-16	...	2015-12-17	2015-12-18	2015-12-21	2015-12-22	2015-12-23	2015-12-24	2015-12-28	2015-12-29	2015-12-30	2015-12-31
EmployeeID
1	2015-01-02 16:56:15	2015-01-05 17:20:11	2015-01-06 17:19:05	2015-01-07 16:34:55	2015-01-08 17:08:32	2015-01-09 17:38:29	2015-01-12 16:58:39	2015-01-13 18:02:58	2015-01-15 17:22:13	2015-01-16 17:35:11	...	0	0	2015-12-21 17:15:50	2015-12-22 17:27:51	2015-12-23 16:44:44	2015-12-24 17:47:22	2015-12-28 18:00:07	2015-12-29 17:22:30	2015-12-30 17:40:56	2015-12-31 17:17:33
2	2015-01-02 18:22:17	2015-01-05 17:48:22	0	2015-01-07 17:09:06	2015-01-08 17:34:04	2015-01-09 16:52:29	2015-01-12 17:36:48	2015-01-13 18:00:13	2015-01-15 17:14:44	2015-01-16 17:40:57	...	2015-12-17 17:10:50	2015-12-18 18:31:28	2015-12-21 17:34:16	2015-12-22 18:16:35	2015-12-23 17:38:18	0	2015-12-28 17:08:38	2015-12-29 17:54:46	2015-12-30 18:31:35	2015-12-31 17:40:58
3	2015-01-02 16:59:14	2015-01-05 17:06:46	2015-01-06 16:38:32	2015-01-07 16:33:21	2015-01-08 17:24:22	2015-01-09 16:57:30	2015-01-12 17:28:54	2015-01-13 17:21:25	2015-01-15 17:21:29	2015-01-16 17:18:13	...	2015-12-17 17:06:23	2015-12-18 17:02:23	2015-12-21 17:20:17	2015-12-22 16:32:50	2015-12-23 16:59:43	2015-12-24 16:58:25	2015-12-28 16:43:31	2015-12-29 17:09:56	2015-12-30 17:06:25	2015-12-31 17:15:50

	2015-01-02	2015-01-05	2015-01-06	2015-01-07	2015-01-08	2015-01-09	2015-01-12	2015-01-13	2015-01-15	2015-01-16	...	2015-12-17	2015-12-18	2015-12-21	2015-12-22	2015-12-23	2015-12-24	2015-12-28	2015-12-29	2015-12-30	2015-12-31
EmployeeID
1	7.208333	7.189722	7.410833	7.006667	7.289722	7.484444	7.262778	7.831111	7.346944	7.267500	...	0.000000	0.000000	7.339167	7.395833	6.504722	7.596389	7.773889	7.315000	7.778889	7.080278
2	8.109167	7.454722	0.000000	7.396944	7.416667	7.150833	7.611389	7.278889	7.613056	7.727500	...	7.928333	7.903056	7.753889	7.712222	7.435556	0.000000	7.614722	7.982500	7.986111	8.227222
3	6.692500	7.265556	6.405278	6.765000	7.345000	6.861389	7.418611	6.999722	7.438333	7.210278	...	7.218333	6.785833	7.163611	6.801667	6.730278	6.849722	7.023889	7.438889	7.538889	6.786389
4	7.338333	7.291944	6.943056	6.919444	6.850833	7.193056	6.998611	7.306389	6.876667	6.907778	...	7.455833	7.629167	6.846667	7.326389	7.413611	7.085000	7.447222	7.416667	7.366389	7.133056
5	8.055556	7.988056	7.682500	7.806111	7.662222	7.721667	8.365000	8.257222	8.260000	8.403611	...	8.320000	7.903611	7.665000	7.957500	7.786944	8.249444	7.662222	8.268611	7.953333	8.018056

	Age	DistanceFromHome	Education	JobLevel	MonthlyIncome	NumCompaniesWorked	PercentSalaryHike	StockOptionLevel	TotalWorkingYears	TrainingTimesLastYear	...	JobRole_Manufacturing Director	JobRole_Research Director	JobRole_Research Scientist	JobRole_Sales Executive	JobRole_Sales Representative	MaritalStatus_Divorced	MaritalStatus_Married	MaritalStatus_Single	Gender_Female	Gender_Male
0	1.541369	-0.393938	-0.891688	-0.961486	1.405136	-0.678464	-1.150554	-0.932014	-1.322079	2.483396	...	-0.330808	-0.239904	-0.497873	-0.533821	-0.244625	-0.534873	1.088232	-0.685565	1.224745	-1.224745
1	-0.648668	0.099639	-1.868426	-0.961486	-0.491661	-1.079486	2.129306	0.241988	-0.678877	0.155707	...	-0.330808	-0.239904	2.008543	-0.533821	-0.244625	-0.534873	-0.918921	1.458650	1.224745	-1.224745
2	-0.539166	0.963398	1.061787	1.749610	2.725053	-0.678464	-0.057267	2.589994	-0.807517	-0.620189	...	-0.330808	-0.239904	-0.497873	1.873287	-0.244625	-0.534873	1.088232	-0.685565	-0.816497	0.816497
3	0.117845	-0.887515	2.038524	0.845911	0.386301	0.123580	-1.150554	2.589994	0.221606	1.707500	...	-0.330808	-0.239904	-0.497873	-0.533821	-0.244625	-0.534873	1.088232	-0.685565	-0.816497	0.816497
4	-0.539166	0.099639	-1.868426	-0.961486	-0.884109	0.524602	-0.877232	1.415991	-0.292956	-0.620189	...	-0.330808	-0.239904	-0.497873	1.873287	-0.244625	-0.534873	-0.918921	1.458650	-0.816497	0.816497
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
4405	0.555852	-0.517332	1.061787	-0.961486	-0.100700	0.123580	0.489376	0.241988	-0.164315	1.707500	...	-0.330808	-0.239904	2.008543	-0.533821	-0.244625	-0.534873	-0.918921	1.458650	1.224745	-1.224745
4406	-0.867672	-0.887515	1.061787	-0.961486	-0.812504	-0.277442	-0.057267	-0.932014	-0.164315	-0.620189	...	-0.330808	-0.239904	-0.497873	-0.533821	-0.244625	1.869602	-0.918921	-0.685565	-0.816497	0.816497
4407	-1.305679	1.950552	-0.891688	-0.057788	-0.595138	-1.079486	1.309341	-0.932014	-0.807517	0.931603	...	-0.330808	-0.239904	-0.497873	1.873287	-0.244625	-0.534873	1.088232	-0.685565	-0.816497	0.816497
4408	0.555852	1.086793	-0.891688	-0.961486	-0.872210	-1.079486	-0.330589	0.241988	-0.164315	-0.620189	...	-0.330808	-0.239904	-0.497873	-0.533821	-0.244625	1.869602	-0.918921	-0.685565	-0.816497	0.816497
4409	0.336849	2.320735	0.085049	-0.057788	-0.219901	-1.079486	-0.877232	-0.932014	-0.164315	2.483396	...	-0.330808	-0.239904	-0.497873	-0.533821	-0.244625	1.869602	-0.918921	-0.685565	-0.816497	0.816497

	Age	Attrition	BusinessTravel	Department	DistanceFromHome	Education	EducationField	EmployeeCount	EmployeeID	Gender	...	NumCompaniesWorked	Over18	PercentSalaryHike	StandardHours	StockOptionLevel	TotalWorkingYears	TrainingTimesLastYear	YearsAtCompany	YearsSinceLastPromotion	YearsWithCurrManager
0	51	No	Travel_Rarely	Sales	6	2	Life Sciences	1	1	Female	...	1.0	Y	11	8	0	1.0	6	1	0	0
1	31	Yes	Travel_Frequently	Research & Development	10	1	Life Sciences	1	2	Female	...	0.0	Y	23	8	1	6.0	3	5	1	4
2	32	No	Travel_Frequently	Research & Development	17	4	Other	1	3	Male	...	1.0	Y	15	8	3	5.0	2	5	0	3
3	38	No	Non-Travel	Research & Development	2	5	Life Sciences	1	4	Male	...	3.0	Y	11	8	3	13.0	5	8	7	5
4	32	No	Travel_Rarely	Research & Development	10	1	Medical	1	5	Male	...	4.0	Y	12	8	2	9.0	2	6	0	4

	count	unique	top	freq	mean	std	min	25%	50%	75%	max
Age	4410.0	NaN	NaN	NaN	36.92381	9.133301	18.0	30.0	36.0	43.0	60.0
Attrition	4410	2	No	3699	NaN	NaN	NaN	NaN	NaN	NaN	NaN
BusinessTravel	4410	3	Travel_Rarely	3129	NaN	NaN	NaN	NaN	NaN	NaN	NaN
Department	4410	3	Research & Development	2883	NaN	NaN	NaN	NaN	NaN	NaN	NaN
DistanceFromHome	4410.0	NaN	NaN	NaN	9.192517	8.105026	1.0	2.0	7.0	14.0	29.0
Education	4410.0	NaN	NaN	NaN	2.912925	1.023933	1.0	2.0	3.0	4.0	5.0
EducationField	4410	6	Life Sciences	1818	NaN	NaN	NaN	NaN	NaN	NaN	NaN
EmployeeCount	4410.0	NaN	NaN	NaN	1.0	0.0	1.0	1.0	1.0	1.0	1.0
Gender	4410	2	Male	2646	NaN	NaN	NaN	NaN	NaN	NaN	NaN
JobLevel	4410.0	NaN	NaN	NaN	2.063946	1.106689	1.0	1.0	2.0	3.0	5.0
JobRole	4410	9	Sales Executive	978	NaN	NaN	NaN	NaN	NaN	NaN	NaN
MaritalStatus	4410	3	Married	2019	NaN	NaN	NaN	NaN	NaN	NaN	NaN
MonthlyIncome	4410.0	NaN	NaN	NaN	65029.312925	47068.888559	10090.0	29110.0	49190.0	83800.0	199990.0
NumCompaniesWorked	4410.0	NaN	NaN	NaN	2.691837	2.493912	0.0	1.0	2.0	4.0	9.0
Over18	4410	1	Y	4410	NaN	NaN	NaN	NaN	NaN	NaN	NaN
PercentSalaryHike	4410.0	NaN	NaN	NaN	15.209524	3.659108	11.0	12.0	14.0	18.0	25.0
StandardHours	4410.0	NaN	NaN	NaN	8.0	0.0	8.0	8.0	8.0	8.0	8.0
StockOptionLevel	4410.0	NaN	NaN	NaN	0.793878	0.851883	0.0	0.0	1.0	1.0	3.0
TotalWorkingYears	4410.0	NaN	NaN	NaN	11.277324	7.77449	0.0	6.0	10.0	15.0	40.0
TrainingTimesLastYear	4410.0	NaN	NaN	NaN	2.79932	1.288978	0.0	2.0	3.0	3.0	6.0
YearsAtCompany	4410.0	NaN	NaN	NaN	7.008163	6.125135	0.0	3.0	5.0	9.0	40.0
YearsSinceLastPromotion	4410.0	NaN	NaN	NaN	2.187755	3.221699	0.0	0.0	1.0	3.0	15.0
YearsWithCurrManager	4410.0	NaN	NaN	NaN	4.123129	3.567327	0.0	2.0	3.0	7.0	17.0
MeanTimeWorkOverYear2015	4410.0	NaN	NaN	NaN	6.839664	1.320894	4.94	5.78	6.55	7.4575	10.47
AbsenceOverYear2015	4410.0	NaN	NaN	NaN	12.734694	5.503779	1.0	8.0	13.0	17.0	24.0
JobInvolvement	4410.0	NaN	NaN	NaN	2.729932	0.7114	1.0	2.0	3.0	3.0	4.0
PerformanceRating	4410.0	NaN	NaN	NaN	3.153741	0.360742	3.0	3.0	3.0	3.0	4.0
EnvironmentSatisfaction	4410.0	NaN	NaN	NaN	2.72517	1.089852	1.0	2.0	3.0	4.0	4.0
JobSatisfaction	4410.0	NaN	NaN	NaN	2.729478	1.098904	1.0	2.0	3.0	4.0	4.0
WorkLifeBalance	4410.0	NaN	NaN	NaN	2.763492	0.703541	1.0	2.0	3.0	3.0	4.0

	Age	Attrition	DistanceFromHome	Education	JobLevel	MonthlyIncome	NumCompaniesWorked	PercentSalaryHike	StockOptionLevel	TotalWorkingYears	...	JobRole_Manufacturing Director	JobRole_Research Director	JobRole_Research Scientist	JobRole_Sales Executive	JobRole_Sales Representative	MaritalStatus_Divorced	MaritalStatus_Married	MaritalStatus_Single	Gender_Female	Gender_Male
EmployeeID
1	51	No	6	2	1	131160	1.0	11	0	1.0	...	0	0	0	0	0	0	1	0	1	0
2	31	Yes	10	1	1	41890	0.0	23	1	6.0	...	0	0	1	0	0	0	0	1	1	0
3	32	No	17	4	4	193280	1.0	15	3	5.0	...	0	0	0	1	0	0	1	0	0	1
4	38	No	2	5	3	83210	3.0	11	3	13.0	...	0	0	0	0	0	0	1	0	0	1
5	32	No	10	1	1	23420	4.0	12	2	9.0	...	0	0	0	1	0	0	0	1	0	1

	feature1	feature2	Correlation
2114	Gender_Male	Gender_Female	1.00
1174	Department_Sales	Department_Research & Development	0.91
562	YearsWithCurrManager	YearsAtCompany	0.77
742	PerformanceRating	PercentSalaryHike	0.77
1033	BusinessTravel_Travel_Rarely	BusinessTravel_Travel_Frequently	0.75

	Algorithmes	precision_train	recall_train	f1_train
0	Logisitic Regression	0.619289	0.214035	0.318123
1	KNN	0.583893	0.305263	0.400922
2	RandomForest	0.991786	0.847368	0.913907
3	Perceptron	0.355049	0.382456	0.368243
4	Descente de gradient stochastique	0.415254	0.257895	0.318182

	Algorithmes	Best_param	Best_estimator	Cross-Validation score Train
0	Logisitic Regression	{'penalty': 'none', 'solver': 'newton-cg'}	LogisticRegression(penalty='none', solver='new...	0.852890
1	KNN	{'metric': 'manhattan', 'n_neighbors': 4, 'wei...	KNeighborsClassifier(metric='manhattan', n_nei...	0.984691
2	RandomForest	{'bootstrap': False, 'max_depth': 40, 'max_fea...	(DecisionTreeClassifier(max_depth=40, max_feat...	0.988378
3	Perceptron	{'fit_intercept': True, 'max_iter': 20, 'penal...	Perceptron(max_iter=20, penalty='l1')	0.796460
4	Descente de gradient stochastique	{'early_stopping': True, 'learning_rate': 'opt...	SGDClassifier(early_stopping=True, loss='perce...	0.767577

	Analyse	Jeu d'entrainement	Jeu de test	Différence
0	precision_score	0.613065	0.571429	0.041637
1	recall_score	0.214035	0.198582	0.015454
2	f1_score	0.317295	0.294737	0.022558

	Analyse	Jeu d'entrainement	Jeu de test	Différence
0	precision_score	0.960894	0.934783	0.026111
1	recall_score	0.905263	0.304965	0.600299
2	f1_score	0.932249	0.459893	0.472356

	Analyse	Jeu d'entrainement	Jeu de test	Différence
0	precision_score	0.996063	0.772727	0.223336
1	recall_score	0.887719	0.361702	0.526017
2	f1_score	0.938776	0.492754	0.446022

	Analyse	Jeu d'entrainement	Jeu de test	Différence
0	precision_score	0.338798	0.367521	-0.028724
1	recall_score	0.326316	0.304965	0.021351
2	f1_score	0.332440	0.333333	-0.000894

	MeanTimeWorkOverYear2015	AbsenceOverYear2015
EmployeeID
1	6.43	17
2	6.79	13
3	6.30	7
4	6.36	14
5	7.35	4

	Analyse	Jeu d'entrainement	Jeu de test	Différence
0	precision_score	0.315690	0.327869	-0.012179
1	recall_score	0.292982	0.283688	0.009295
2	f1_score	0.303913	0.304183	-0.000270

CESI HumanForYou¶

Table des matières¶

Préparation de l'environnement¶

Importation des librairies¶

Importation des données¶

Visualisation des données¶

Données du service des ressources humaines¶

Dernière évaluation du manager¶

Enquête qualité de vie au travail¶

Horaires de travail¶

Transformation des données¶

Calcul des durées de travail¶

Concaténation des données¶

Ajout de valeur¶

Suppression de colonne¶

Normalisation¶

Standardisation¶

Ethique¶

Création des jeux de données¶

Analyses statistiques¶

Analyse de l'attrition¶

Courbes de densité superposées¶

Boîtes à moustache¶

Histogrammes¶

Analyste des données entrantes¶

Histogramme¶

Corrélation des données¶

Conclusion des analyses¶

Algorithmes¶

Evaluation des modèles¶

Première évaluation¶

Hyperparameter tuning¶

Param reglog¶

Param knn¶

Param random_forest :¶

Param preceptron :¶

Param descente de gradient stochastique (mini-batch size) :¶

Regression hyperparameter tuning¶

KNNeighboors hyperparameter tuning¶

Random Forest hyperparameter tuning¶

Perceptron hyperparameter tuning¶

Descente de Gradient Stochastique hyperparameter tuning¶

Résumé¶

Deuxième évaluation¶

Regression logistique¶

KNN¶

Random Forest¶

Perceptron¶

Descente de gradient stochastique¶

Comparaison des courbes ROC¶

Conclusion¶

Etude des features¶

MeanTimeWorkOverYear2015¶

MonthlyIncome¶

DistanceFromHome¶

Pistes à suivre¶